SpheroSeg – advancing tumor spheroid analysis with deep learning

We are pleased to share our new publication in Computer Methods and Programs in Biomedicine, presenting SpheroSeg, an open-source platform for bright-field segmentation and quantitative analysis of tumor spheroids. SpheroSeg was developed thanks to the collaboration between biologists from our university and computer scientists from the Institute of Information Theory and Automation of the Czech Academy of Sciences.

Three-dimensional tumor spheroids are valuable in vitro models for studying tumor growth, invasion, and, most importantly, treatment effect. However, their quantitative analysis remains challenging due to variations in imaging conditions and annotation policies across experiments and datasets.

For this reason, we developed SpheroHQ, a dataset containing 22,683 expert-corrected images of spheroids from seven cancer cell lines. Combined with external spheroid datasets, it forms the 32,367-image SpheroMix.

The study also provides a systematic benchmark of eight deep-learning segmentation architectures. A stratified evaluation protocol was used to distinguish within-dataset performance, annotation-policy effects, and true out-of-distribution generalization.

Beyond the dataset and benchmark, SpheroSeg provides a deployable web-based analysis workflow supporting GPU-accelerated spheroid segmentation, polygon-based correction, morphometric measurements, and data export for downstream analysis. Three complementary models are available: CBAM-ResUNet, SegFormer-B0, and MambaBot-UNet. The platform is also available as a Dockerized application.

SpheroSeg therefore brings together high-quality experimental data, reproducible benchmarking, and an accessible analysis platform, while highlighting an important methodological point: strong performance on data from one experimental setting does not necessarily translate into equally strong generalization to different imaging and annotation conditions.

And what does one of the authors, Dr. S. Rimpelová, say? "From sharing a desk at an eight-year grammar school to becoming co-corresponding authors – years later, with Dr. A. Novozámský, we brought together our worlds of biology and IT, resulting in SpheroSeg."

Reference

SpheroSeg: Advancing tumor spheroid analysis through open-source deep learning in Computer Methods and Programs in Biomedicine, doi: 10.1016/j.cmpb.2026.109602